Research

Everything in this section in one place — newest stories first.

LatestResearch

Two New Preprints Tackle Oncology AI's Blind Spots: From Tumor Scoring to Hospital-Ready Orchestration

A June 2026 preprint (arXiv:2606.27579) introduces a two-stage MIL framework using zero-inflated beta modeling to predict tumor proportion scores in lung cancer pathology slides, outperforming standard regression baselines.

ResearchJuly 15, 2026

When AI Draws the Map, the Territory Changes: A Study on Classifying Public Administration Research

A new arXiv paper tests five methods, including author-defined, citation-based, and AI-assisted approaches, for classifying public administration scholarship using Web of Science and OpenAlex.

Developer ToolsJuly 15, 2026

Plainbook Reimagines the Notebook for Scientists Who Don't Write Code

Plainbook is a research notebook system that generates code automatically from natural language cell descriptions, treating the description rather than the code as the persistent artifact.

ResearchJuly 15, 2026

Decade-Long Review of AI in STEM Education Charts a Shift From Tutoring Bots to Inquiry-Driven Learning

A bibliometric review of 242 papers (2015-2025) on AI in STEM education was posted to arXiv on June 10, 2026, and accepted at the ISLS26 conference.

ResearchJuly 15, 2026

Audit Finds Most 'Risk-Aware' Claims in Distributional RL Agents Don't Hold Up

An arXiv audit of QR-DQN, C51, and IQN agents on MinAtar found 40-95% of top risk-related claims were statistically refuted at 95% confidence

ResearchJuly 15, 2026

New Framework Cuts Decoder Compute 32x for 5G Channel Feedback Compression

A new framework embeds Contrastive Predictive Coding into 3GPP CSI compression, cutting decoder compute by 32x while keeping over 90% reconstruction accuracy.

LLMsJuly 15, 2026

EU Researchers Chart a Domain-Specific Path for LLMs in the Social Sciences and Humanities

European project LLMs4EU, part of the ALT-EDIC infrastructure, has developed a domain-adaptive LLM framework for Social Sciences and Humanities research.

ResearchJuly 15, 2026

What 11 Studies Reveal About AI's Real Impact on English Learners in Arab Universities

A systematic review of AI tools for English learners in Arab universities filtered 184 candidate studies down to just 11 meeting strict empirical criteria.

AI AgentsJuly 15, 2026

Multi-Agent Reinforcement Learning Squeezes 18% More Profit From Dairy Farm Batteries

A July 2026 arXiv paper applies multi-agent reinforcement learning to battery management on Irish dairy farms, lifting energy arbitrage profits by up to 18% over rule-based systems.

ResearchJuly 15, 2026

New Distance Metric Tackles a Long-Standing Blind Spot in Categorical Data Clustering

A paper on learnable distance weighting for categorical data clustering, originally published in IEEE TPAMI in 2022, was reposted to arXiv on July 6, 2026.

ResearchJuly 15, 2026

Two New Systems Push AI Toward Autonomous Science Writing and Long-Form Story Memory

Prompt-to-Paper, detailed in a July 5, 2026 arXiv paper, generates submission-ready scientific manuscripts for about $0.31 each, using an autonomous coding agent to run real computational biology experiments instead of synthetic results.

LLMsJuly 15, 2026

A New Format for Chain-of-Thought Could Cut LLM Reasoning Costs by 85%

Researchers introduce Structured Thoughts, a framework that formats LLM chain-of-thought reasoning into alternating <try> and <outcome> blocks

ResearchJuly 15, 2026

AdaStop Cuts DNN Testing Costs by Knowing When to Stop Looking for Bugs

AdaStop is a new framework that treats DNN fault testing as a cost-benefit decision, stopping labeling once the marginal fault discovery rate drops below a defined threshold.

AI AgentsJuly 15, 2026

Multi-Agent AI System Hits 98.6% Accuracy Reading Gastric Biopsy Reports in Singapore Study

Nimblemind's nMAS multi-agent AI system hit 98.61% accuracy across 216 feature-case decisions on gastric biopsy pathology reports from Singapore

ResearchJuly 15, 2026

Physics-Informed Neural Networks Learn to Predict Shockwaves Without Rerunning Simulations

A physics-informed neural network (PINN) now models how shockwaves propagate through steel-aluminum bimaterial systems, built around the governing equations of linear elasticity rather than raw data patterns alone.

EducationJuly 15, 2026

Students Are Skipping the Fundamentals of AI, and It's Skewing How They Learn It

A study of 158 students, academics, and staff found students often master advanced GenAI tasks before understanding foundational concepts, an 'inverted' skill profile.

LLMsJuly 15, 2026

Nadella's Warning: Companies Using AI Are 'Paying Twice'

Satya Nadella says enterprises using AI models pay both in token costs and by leaking proprietary data back to model providers.

AI AgentsJuly 15, 2026

CHARLIE Brings Multi-Agent RAG to Digital Forensics, Keeping Evidence Off the Cloud

CHARLIE is a multi-agent RAG system built specifically for digital forensic investigations, detailed in an arXiv paper submitted July 1, 2026.

All articles